SGER: Advances in Difficult Aspects of Reference Resolution

SGER：指称解析难点方面的进展这个探索性项目正在研究指称解析领域的两个困难现象：(1)复数代词的解析，这些复数代词既可以与一个句法成分相关联，也可以与文本中不同地方的一组成分相关联；他们，那两个，所有三个，等等(2)第三人称单数和指示代词的解析，它们既可以指代一个句法成分，也可以指代一个文本跨度：it， that, this, all of this，等等。迄今为止，这些现象的处理方式有限，因为最广泛用于指称解析的方法都期望指称表达式的先行词是前面上下文中的单个名词短语。在所研究的案例中，先行词不需要以这种方式加以限制。这项工作旨在实现类似人类的智能，以支持智能代理和人之间的复杂通信。它在理论上是有根据的，但把实际问题放在首位。组件算法可以在相对较短的时间内为任何系统的任何领域提供参考分辨率，推进已知类型的技术，而不依赖于任何意想不到的突破。分发给社区的资源包括独立于系统的参考解析算法、相关示例及其语义表示的语料、参考解析引擎，以及对成就、突出问题和进一步发展路径的重要调查。

SGER: Advances in Difficult Aspects of Reference ResolutionThis exploratory project is investigating two difficult phenomena in the domain of reference resolution: (1) the resolution of plural pronouns that can corefer either with a single syntactic constituent or with a set of constituents from different places in the text: they, those two, all three, etc. and (2) the resolution of 3rd person singular and demonstrative pronouns that can corefer either with a single syntactic constituent or with a span of text: it, that, this, all of this, etc. These phenomena have been treated in limited ways to date because the methods that have been most widely deployed for reference resolution have expected the antecedent for referring expressions to be a single noun phrase in the preceding context. In the cases under study, the antecedent need not be constrained in this way. This work is aimed at achieving human-like intelligence to support, for example, sophisticated communication between intelligent agents and people. It is theoretically grounded but keeps practical concerns at the fore. The component algorithms can inform reference resolution in any domain by any system in the relatively near term, advancing known types of technology and not relying on any unexpected breakthroughs. Among the resources to be distributed to the community are system-independent reference resolution algorithms, a corpus of relevant examples and their semantic representations, a reference resolution engine, and a critical survey of accomplishments, outstanding issues, and paths for further development.